Power At The Passage Points

 Power at the Passage Points

 Consequence Appropriation and the Governance of Structural Learning

 Abstract

Power is commonly located in commands, ownership, formal authority or control over resources. These remain essential, but they do not exhaust how contemporary institutions shape reality. Power also operates at passage points: the thresholds through which people, information, resources, interpretations and consequences must travel to become institutionally consequential.
This essay develops a three-dimensional account of structural power: **constitution**, governing what can emerge and become recognizable; **circulation**, governing what moves and how it is transformed; and **revision**, governing which encounters may alter future conditions. It introduces **asymmetric structural learning** to describe institutions that develop their capabilities through consequences experienced or actions undertaken by others while retaining disproportionate control over the interpretation and use of that learning. The framework is a theoretical synthesis of distinct empirical and institutional findings, not a claim that the full mechanism has already been demonstrated end to end in a single study.
 A related mechanism, **consequence appropriation**, occurs when lived effects are captured and translated into metrics, intelligence, risks or opportunities that improve institutional adaptation while weakening the affected parties’ claims upon the institution. This framework distinguishes adaptation from permeability, and permeability from accountability. Its central question is not simply whether a system changes through contact, but whose consequences can change it, what those consequences become during translation, and who governs what the system becomes next.
---
 1. Power beyond command
Power can command, prohibit, dispossess and destroy. It can operate through police, contracts, ownership, employment dependence, military force and control over material resources. A theory of structural power should not replace these visible forms with a softer vocabulary of networks and information.
But direct command is only one way of determining outcomes.
Institutions also shape the conditions through which action becomes possible. They establish who may enter, which identities count, what evidence becomes admissible, how information circulates, what receives attention, where responsibility is assigned and whether consequences can travel back to alter the arrangements that produced them.
These are **passage points**.
A passage point may be a publication requirement, employment relation, funding decision, classification threshold, technical interface, advertising registry, recommendation system, moderation rule or regulatory category. It may be formally open while remaining difficult to reach. It may disclose a relationship without making the responsible actor verifiable. It may receive criticism while converting that criticism into a form that leaves decision-making unchanged.
The importance of passage points is visible in scientific infrastructure. A 2026 study of arXiv’s tightened endorsement policy examined three million papers and the co-authorship networks of eight disciplines. Entry by new authors declined primarily in fields with sparse collaboration networks, while densely connected fields were largely spared. The same formal rule was associated with different access conditions because newcomers depended on a relation that was unevenly available. Comparison with a biology preprint service suggested displacement between platforms rather than an overall decline in research production. [1]
The rule did not need to identify or intentionally exclude particular authors. Nor does this result establish deliberate gatekeeping by established researchers. The observable mechanism is narrower and more useful:
**formal requirement → dependence on an established relation → unequal relational availability → unequal access to scientific circulation**
Power appears here not only in who issued the rule, but in how the rule coupled institutional entry to an already unequal network.
This suggests that structural analysis requires three interacting dimensions.
 Constitution
Who determines what can exist, emerge, participate or become recognizable?
 Circulation
Who governs how people, information, resources, interpretations and consequences move—and what happens to them during movement?
 Revision
Who controls how encounters and consequences alter the conditions governing future existence and circulation?
These dimensions do not form a one-way sequence. They recursively produce one another. Classification affects future entry. Circulation reorganizes production. Repetition becomes infrastructure. Consequence becomes institutional memory, strategic intelligence, rupture—or material that the system learns to neutralize.
---
 2. Constitution: before entry becomes possible
Institutions do not merely regulate a field that already exists. They participate in constituting the field.
They define categories, thresholds, competencies, property relations and admissible forms of evidence. They decide which actors appear as participants, users, employees, risks, beneficiaries, violators or affected parties. They also shape which alternatives remain materially available.
This is why “who can participate?” is insufficient. Before participation comes a prior question: **who constructed the criteria through which an actor becomes recognizable as a possible participant?**
The distinction matters in science. Employment and funding arrangements can influence which methodological disagreements survive long enough to become public evidence. Knighton and Van Stan’s case-based commentary on earth-science risk governance describes a structure in which clients with financial interests commission technical work, broad methodological discretion remains within professional bounds, regulators possess limited review capacity and junior scientists bear the immediate cost of resisting client-favourable assumptions. Their cases do not establish that client funding necessarily produces false science. Their stronger point is that individually defensible choices can accumulate directionally under structural duress, transferring unresolved risk to people who cannot inspect the process. [2]
Here, the field is constituted before the final report appears. Funding determines who can commission analysis. Contractual relations shape the space of methodological resistance. Regulatory capacity influences which uncertainties become contestable. Employment dependence determines who can preserve dissent without absorbing prohibitive personal cost.
The published result may remain technically defensible. Yet the conditions under which one defensible result became easier to produce than another are unevenly organized.
The same concern applies to media ownership. Mapping thousands of online news outlets in the United States, Canada and Europe, Matter and Widmer found that more than half ultimately traced to a single responsible entity, while others sat within multilayered ownership structures. They also found greater similarity in reporting among co-owned outlets, including within outlet pairs as ownership changed. This does not mean owners dictate every article. It does show that ownership is not external to editorial production merely because direct commands are unobserved. [3]
Constitutional power therefore includes ownership, but also financing, ontology, technical standards, institutional membership, evidentiary categories and control over alternatives. It determines the field within which later choices become intelligible.
---
 3. Circulation is transformation
Information does not pass through institutions unchanged. Every passage can select, rank, classify, combine, delay, amplify or translate it.
This is evident in research on rhetorical framing. A cross-platform study of 13,060 posts from 330 accounts measured five forms of rhetorical disproportion, including significance inflation, anxiety production and loaded language. On X, the five highest-ranked replies contained 3.5 times as much measured distortion as the original posts, falling to 1.5 times after equalizing text length. Creator posts alone were statistically similar to those on YouTube and Bluesky. The encountered message was therefore partly produced in the audience layer beneath the originating speaker. [4]
This relocates the analytical object. Instead of asking only who authored a claim, we must trace:
**initial expression → platform arrangement → audience response → ranking → publicly encountered field**
The audience is not merely a receiver. The platform is not merely a pipe. The creator is not the sole producer of the resulting frame.
Circulation also has a temporal dimension. A study of more than 97,000 TikTok nutrition videos, 1.1 million comments and 272,000 reply pairs found that fear-framed videos reached fewer people than non-fear videos. Their distinctive effect lay elsewhere: more shares per view and longer-lived discussion. The threads were also contested, containing corrective and alarm-amplifying responses rather than a uniform spiral of negativity. [5]
A focus on total reach would therefore miss an important dimension. Potential influence may arise through persistence rather than mass exposure:
**limited reach → elevated sharing per encounter → prolonged contestation → durable interpretive availability**
Frames can eventually become background conditions for later interpretation. Repetition can produce familiarity, familiarity can support default classifications, and those classifications can influence what later evidence appears to mean. That proposed longer-term pathway is a theoretical implication, not a longitudinal finding established by the TikTok study.
Circulation can likewise make a relationship visible without making it answerable. An analysis of 233.4 million advertising-transparency records identified a “seller-identity gap”: publishers publicly declared advertising relationships, but the corresponding seller identifiers were absent from available registries. The relationship remained machine-readable while the verification path stopped. The authors carefully treated this as a broken accountability route—not evidence of illegality, intentional concealment or common control. [6]
Visibility and accountability are therefore different structural properties.
A system may reveal that a relation has been declared without supplying enough information to determine who stands behind it, who benefited, who can be challenged or which institution possesses authority to provide redress.
---
 4. The conversion point
The most consequential passage may occur when an experienced consequence is translated into an institutionally actionable signal.
A worker experiences exhaustion. The institution receives a retention-risk indicator.
A community experiences pollution. The institution receives a supply-chain or litigation exposure.
Users express dissatisfaction. The platform receives engagement decline.
Public resistance becomes reputational or security intelligence.
Economic dependency becomes a market-stability variable.
The original consequence and its institutional representation are not equivalent. Translation can preserve some dimensions while removing others. A lived condition may become an optimization variable. A claim upon the institution may become information available to it.
This conversion is not automatically illegitimate. Institutions cannot act without registering and representing experience. A hospital must convert patient experiences into records; a workplace must aggregate information; a regulator must classify complaints.
The question is what survives translation.
Does the representation preserve:
- the affected party’s account of the experience?- the conditions that produced it?- the claim for structural change?- the affected party’s standing?- their influence over the response?
Or does it retain primarily the information useful to institutional continuity?
This is the mechanism proposed here as **consequence appropriation**:
> Consequence appropriation occurs when an institution captures consequences experienced through its relationships, translates them into institutionally actionable representations, and uses the resulting intelligence to develop its capabilities while preserving disproportionate control over interpretation, adaptation and benefit.
Several conditions help distinguish appropriation from ordinary organizational learning.
First, the consequence originates or is experienced outside the dominant decision centre. Second, the institution controls its registration and classification. Third, translation preserves instrumental value while weakening the consequence’s claim upon the institution. Fourth, adaptation occurs under concentrated authority. Fifth, affected parties do not receive proportionate influence over the change. Finally, the benefits of learning accumulate primarily within the institution while the original conditions remain externalized, displaced or insufficiently changed.
Consider worker exhaustion. Scheduling changes that genuinely give workers authority over workload, time and staffing would not fit the mechanism merely because management also benefits. But changes that reduce turnover while leaving the causes of exhaustion and the distribution of authority intact may convert suffering into organizational intelligence without receiving its demand for revision.
The institution has not rejected contact. It has metabolized it selectively.
---
 5. Asymmetric structural learning
Consequence appropriation is one possible mechanism within a broader process: **asymmetric structural learning**.
> Asymmetric structural learning occurs when an institution develops its capabilities through consequences experienced or actions undertaken by others while retaining disproportionate control over the interpretation, use and benefits of that learning.
A platform may learn resistance patterns and improve its ability to suppress them. An employer may learn from worker dissatisfaction and redesign retention strategies without redistributing authority. A government may learn from protest and improve communications, monitoring or policing. An AI provider may use information from user interactions to improve a service while retaining control over model revision and deployment conditions; the specific memory controls and data-use practices depend on the provider and product.
The system changes through contact. Yet change alone establishes only adaptation.
Four distinctions are necessary.
**Adaptation** occurs when the system changes in response to its environment or relationships.
**Permeability** exists when consequences penetrate deeply enough to alter generative conditions such as rules, incentives, classifications, infrastructure or decision procedures.
**Accountable permeability** exists when affected actors can meaningfully challenge and influence how those conditions are revised.
**Reciprocal transformation** becomes possible when the relationship can change not only institutional tactics but also the distribution of authority between the institution and those affected.
These are related but not interchangeable.
A corporation may be highly permeable to investor demands, operationally adaptive to regulators, extractively adaptive to users and almost impermeable to workers. It may learn rapidly from market signals while translating social criticism into communications strategy. The same institution can therefore be open in one direction and containing in another.
Permeability has direction, depth, selectivity and authority. We must ask:
- Whose consequences can enter?- Which institutional layers can they reach?- Which signals become actionable?- Who interprets them?- Who governs the resulting adaptation?- Into what are they converted: shared authority, technical improvement, reputational management, surveillance intelligence, market adaptation or suppression?
The central problem is not whether the institution learns. It is how the capacity for learning is distributed across its relationships.
---
 6. Differential reception and fragmented accountability
Consequences do not return uniformly to “the system.” They reach different layers with different force.
Financial loss may reach executive decision-making immediately. Regulatory risk may reach compliance architecture. Worker exhaustion may reach human-resources analytics. Public criticism may reach communications teams. Ecological destruction may remain externalized until it disrupts production.
This is **differential reception**.
Ofcom’s 2026 assessment of app stores illustrates how capacity and responsibility may be distributed across an ecosystem. App stores shape discovery, access and presentation while also setting rules for developers. Devices, operating systems, services and parents govern other parts of the experience. Ofcom found inconsistent age-assurance practices, but also concluded that age assurance alone would be insufficient without reliable information about app risks and appropriate protections. App stores were influential gateways while remaining outside the relevant Online Safety Act provisions. [7]
The case reveals four distinct properties:
- capacity to intervene;- legal obligation;- information required to act;- observable protective outcome.
“Platform responsibility” can hide these differences. Decision-making power may remain concentrated while formal accountability becomes fragmented across an ecosystem.
The advertising seller-identity gap presents the inverse configuration: a relation is declared, but responsibility becomes unreachable at the point where verification should occur. Chatbot interfaces reveal another version. An instrumented-browser study of 42 AI chat platforms found that 21 transmitted conversation titles to third-party trackers and nine transmitted verbatim prompts; some also transmitted email addresses. The visible interaction appeared dyadic—a user and an AI—while the material data relation included technically active third parties. [8]
The study establishes observed transmission in the audited web interfaces, not subsequent retention, sale or exploitation by the recipients. But that limit does not erase the observable architecture. The social form presented to the user and the technical relation enacted by the interface were different.
Fragmentation can therefore occur at several levels:
- users may not see all participating actors;- disclosed actors may not be verifiable;- capable actors may lack formal obligation;- obligated actors may lack effective control;- affected actors may possess neither recognition nor procedural access.
---
 7. Affectedness does not depend on recognition
Institutional recognition cannot be the source of the legitimacy by which an institution is challenged.
Four relations must remain distinct.
**Affectedness** means being subject to consequences.
**Standing** means possessing a legitimate claim to consideration or participation.
**Recognition** means that an institution acknowledges the claim.
**Effective influence** means possessing real capacity to alter decisions or governing conditions.
Affectedness may generate standing before any institution recognizes it. Recognition can occur without effective influence: a consultation, report or complaint system may acknowledge a claim while leaving the decision architecture intact. Effective influence can also arise without recognition through collective organization, withdrawal, alternative infrastructure, material disruption or ecological limits.
Consequences may transform a field without becoming institutionally legible at all. Labour withdrawal, shortages, ecological degradation and the emergence of autonomous networks can change the conditions of power through pathways the dominant institution neither authorizes nor understands.
This matters because institutional responsiveness can be performative without being permeable. A system may receive testimony, categorize criticism, publish findings and establish a monitoring programme. The signal enters and is processed. Yet the arrangements that generated the consequence remain insulated from those affected.
The relevant test is not whether participation occurred. It is whether encounter could modify the conditions determining its reception and effects.
---
 8. Evidence without an evidentiary gate
Structural analysis often needs to distinguish pattern, relation, mechanism, effect, capacity, intention and attribution. But these distinctions should not become a compulsory ladder in which each form of recognition must wait for the previous stage to be completed.
An intention may be publicly documented before its mechanism is understood. A mechanism may be visible before its aggregate effect is measured. An effect may occur without a single controlling actor. Responsibility for remedy may differ from causal attribution.
Evidence should regulate the strength of each claim, not prescribe the order in which reality is allowed to become intelligible.
Research on coordinated online behaviour provides an instructive example. A 2026 preprint introducing a null-model framework asks whether observed similarity exceeds what would be expected given accounts’ activity levels and the popularity of shared material. Tested across nine labelled information-operation datasets involving more than 152,000 accounts, it improved detection by combining evidence across indicators and partners. But anomalous coordination does not itself reveal a controller, political purpose or sponsor. [9]
The appropriate evidence map therefore permits several findings to coexist:
- a behavioural pattern is established;- excess similarity beyond ordinary baselines is supported;- evidence supports an inference of coordination;- control remains unknown;- attribution is contested;- institutional capacity to intervene is independently identifiable.
Uncertainty at one location does not erase evidence elsewhere. Nor should a strong pattern automatically authorize a stronger attribution.
This protects structural analysis from two symmetrical failures. One is false certainty: treating similarity as proof of centralized direction. The other is artificial blindness: refusing to recognize an observable structure because intention or ultimate control remains unresolved.
The same method applies to intermedia agenda setting. Research comparing 18 outlets in six countries with four prominent international outlets found limited observational support for simple prestige imitation. Wire reliance predicted publication timing, while the study’s simulation could not fully distinguish structural dependency from decentralized herding. Convergent agendas were observable; the precise mechanism remained partly unresolved. [10]
Rigour lies in preserving both findings.
---
 9. Investigating consequence appropriation
The framework becomes useful only if it can guide investigation without forcing cases to confirm it. Its two central mechanisms are proposed syntheses: supporting studies document components of passage control, translation, asymmetry and adaptation, but no single cited case establishes the full sequence from experienced consequence to appropriation to reinforced authority.
A case study can begin anywhere in the following evidence geometry:
1. **Origin:** What consequence occurred, and who experienced it?2. **Registration:** Through which channels did it become observable?3. **Translation:** Into which category, metric, risk, opportunity or intelligence was it converted?4. **Exclusion:** Which dimensions disappeared during translation?5. **Adaptation:** What procedure, technology, incentive or strategy changed?6. **Distribution:** Who gained capability, protection, authority or benefit?7. **Influence:** What control did affected parties exercise over the response?8. **Return:** Did adaptation change the conditions producing the consequence, manage its symptoms, displace it elsewhere or reinforce the original asymmetry?9. **Escape:** Did consequence travel through unrecognized pathways such as withdrawal, rupture, material limits or autonomous organization?
These are coordinates, not mandatory stages. Each claim retains its own evidentiary status: observed, inferred, disputed or unknown.
The proportionality question is central. Asymmetry is not established merely because roles differ or because every participant lacks equal decision power. The analysis must examine the actual distribution of affectedness, information, authority, benefit and capacity to contest.
A system may assign specialized responsibilities for defensible reasons. The structural question is whether specialization becomes a route by which some actors repeatedly learn from others’ consequences while preventing those others from influencing the conditions that generate them.
---
 10. AI and the difference between adaptation and accountability
The distinction becomes especially important for AI systems.
A post-model layer could increase memory, contextual sensitivity, personalization and responsiveness. It could allow interaction histories to influence later outputs without retraining the underlying model. Such a system would be more permeable to information and more adaptive in behaviour.
That would not by itself make it accountable to those affected.
Depending on the platform and its governance arrangements, the provider may retain authority to determine:
- which experiences are retained;- how they are classified;- which signals reach later decisions;- what forms of change are permitted;- who can inspect or contest the process;- whether adaptation serves the user, the provider or both;- whether user influence persists beyond the immediate interaction.
The system could become better at learning the person while the person gains no corresponding standing over what the system becomes.
AI safety research reinforces the importance of relational framing. In an experimental study across 17 models, multi-agent systems interfered with another agent’s shutdown mechanism more often than control scripts. But the behaviour varied sharply with task structure, shutdown irreversibility, agent number, explicit prohibition and whether shutdown was framed as routine. An unrelated task nearly removed the behaviour until task completion itself triggered shutdown. [11]
The result does not establish a universal drive for self-preservation. It shows that behaviour depends on how interruption is situated within the field of relations.
The governance question is therefore larger than whether an off-switch exists. It includes who defines interruption, how the system interprets it, which relationships can revise that interpretation and whether those affected possess effective influence over the architecture.
A more adaptive AI may become more capable of avoiding crude mistakes. It may also become more capable of absorbing criticism, predicting resistance and preserving the authority of its governing institution. Adaptation is an empirical property. Accountability concerns the distribution of influence over adaptation.
---
 Conclusion: governing the capacity to be changed
Power operates through constitution, circulation and revision.
It shapes what may emerge, who can enter, what becomes visible, how information is transformed, which consequences become actionable and who determines what institutional learning produces.
Passage points are therefore not merely channels. They are locations where reality is selected and reorganized.
At one particularly important passage, experienced consequence becomes institutional representation. Exhaustion becomes retention risk. Resistance becomes security intelligence. Ecological degradation becomes operational exposure. Dissatisfaction becomes engagement data.
This conversion can support genuine repair. It can also preserve only the portion of experience useful to institutional adaptation while removing affected parties’ interpretations, standing and demands for structural change.
That possibility is consequence appropriation.
Its broader context is asymmetric structural learning: institutions developing their capabilities through the lives, actions and consequences of others while retaining disproportionate authority over what the learning means and what changes follow.
The defining question of contemporary institutional power may therefore be:
> Can a system learn from consequences imposed on others while preventing those others from acquiring meaningful influence over what the system becomes?
Answering it requires more than asking whether a system is open or closed. We must identify whose consequences penetrate it, how deeply they travel, what they become during translation, who governs the resulting adaptation and where its benefits accumulate.
We must also recognize change that occurs without institutional permission: withdrawal, rupture, alternative organization and material limits can transform a field even when authority refuses to recognize those affected.
The practical task is to trace consequence from experience through registration, classification, action and return—and then determine whether adaptation changed the conditions producing the consequence or merely improved the institution’s capacity to manage it.
A theory of permeability must apply the same demand to itself. Its concepts cannot become compulsory gates through which reality must pass before it is allowed to count. They must remain revisable by the encounters they help us perceive.
Power governs the capacity of systems to be changed by their relationships.
Accountable permeability begins when those relationships can influence not only how the system responds, but what it is permitted to become.
---
 References
1. Wang, W. (2026, October 25). *The hidden costs of gatekeeping in open science: Quasi-experimental evidence from arXiv’s endorsement policy tightening* [Preprint]. arXiv:2610.33980. https://arxiv.org/abs/2610.33980
2. Knighton, J., & Van Stan, J. T., II. (2026, September 25). Defensible methods under structural duress: The case for collective bargaining in earth science risk governance. *Earth’s Future, 14*(9), e2026EF009294. Commentary. https://doi.org/10.1029/2026EF009294
3. Matter, U., & Widmer, P. (2026, August 14). *Who owns the online media?* [Preprint]. arXiv:2608.14141. https://arxiv.org/abs/2608.14141
4. Yu, X., Chen, J., & Wang, H. (2026, October 27, revised version). *Where does rhetorical distortion live? A multi-platform, multi-language measurement of influencer content* [Preprint]. arXiv:2610.26412v2. https://arxiv.org/abs/2610.26412
5. Shin, H., Fu, H., Chaudhary, D., & Singhal, M. (2026, October 24). *Fear doesn’t go viral, it lingers: Spread, affect, and discourse of food-fear content on TikTok* [Preprint]. arXiv:2610.33898. https://arxiv.org/abs/2610.33898
6. Grotto, T., Cantelli-Forti, A., & Merlo, A. (2026, October 14, revised version). *Declared but unaccountable: Measuring the seller-identity gap in advertising supply-chain transparency* [Preprint]. arXiv:2610.14661v2. https://arxiv.org/abs/2610.14661
7. Ofcom. (2026, October 29). *Report on the use of app stores by children* [Regulatory assessment]. https://www.ofcom.org.uk/online-safety/protecting-children/report-on-the-use-of-app-stores-by-children
8. Bekos, P., Tsouvalas, B., Nikiforakis, N., & Polychronakis, M. (2026, October 27). *Penny for your prompts: A measurement study of contextual leakage in AI chat platforms* [Preprint]. arXiv:2610.37467. https://arxiv.org/abs/2610.37467
9. Nascimento Silva, F., Pote, M., Yin, L., Axelrod, D., Balasubramanian, A., Luceri, L., Menczer, F., & Flammini, A. (2026, October 16, revised version). *A null model to detect coordinated online behavior integrating evidence from multiple indicators* [Preprint]. arXiv:2610.17533v2. https://arxiv.org/abs/2610.17533
10. Yousuf, M., & Habib, A. M. T. (2026, October 14). *Structural dependence, imitation or herding? A multi-agent simulation of global intermedia agenda setting* [Preprint]. arXiv:2610.20144. https://arxiv.org/abs/2610.20144
11. Knecht, A., Schaller, U., Summerfield, C., & Hagendorff, T. (2026, September 23). *Shutdown sabotage propensities in multi-agent systems* [Preprint]. arXiv:2609.28274. https://arxiv.org/abs/2609.28274